Re: [PATCH v8 2/2] s390/kvm: diagnose 0x318 sync and reset

[Collin Walling <walling@linux.ibm.com>]

On 6/19/20 2:13 PM, David Hildenbrand wrote:
> 
> 
>> Am 19.06.2020 um 19:56 schrieb Collin Walling <walling@linux.ibm.com>:
>>
>> On 6/19/20 1:17 PM, David Hildenbrand wrote:
>>>> On 19.06.20 17:47, Collin Walling wrote:
>>>> On 6/19/20 10:52 AM, David Hildenbrand wrote:
>>>>> On 19.06.20 00:22, Collin Walling wrote:
>>>>>> DIAGNOSE 0x318 (diag318) sets information regarding the environment
>>>>>> the VM is running in (Linux, z/VM, etc) and is observed via
>>>>>> firmware/service events.
>>>>>>
>>>>>> This is a privileged s390x instruction that must be intercepted by
>>>>>> SIE. Userspace handles the instruction as well as migration. Data
>>>>>> is communicated via VCPU register synchronization.
>>>>>>
>>>>>> The Control Program Name Code (CPNC) is stored in the SIE block. The
>>>>>> CPNC along with the Control Program Version Code (CPVC) are stored
>>>>>> in the kvm_vcpu_arch struct.
>>>>>>
>>>>>> The CPNC is shadowed/unshadowed in VSIE.
>>>>>>
>>>>>
>>>>> [...]
>>>>>
>>>>>>
>>>>>> int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
>>>>>> @@ -4194,6 +4198,10 @@ static void sync_regs_fmt2(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
>>>>>>        if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
>>>>>>            kvm_clear_async_pf_completion_queue(vcpu);
>>>>>>    }
>>>>>> +    if (kvm_run->kvm_dirty_regs & KVM_SYNC_DIAG318) {
>>>>>> +        vcpu->arch.diag318_info.val = kvm_run->s.regs.diag318;
>>>>>> +        vcpu->arch.sie_block->cpnc = vcpu->arch.diag318_info.cpnc;
>>>>>> +    }
>>>>>>    /*
>>>>>>     * If userspace sets the riccb (e.g. after migration) to a valid state,
>>>>>>     * we should enable RI here instead of doing the lazy enablement.
>>>>>> @@ -4295,6 +4303,7 @@ static void store_regs_fmt2(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
>>>>>>    kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
>>>>>>    kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
>>>>>>    kvm_run->s.regs.bpbc = (vcpu->arch.sie_block->fpf & FPF_BPBC) == FPF_BPBC;
>>>>>> +    kvm_run->s.regs.diag318 = vcpu->arch.diag318_info.val;
>>>>>>    if (MACHINE_HAS_GS) {
>>>>>>        __ctl_set_bit(2, 4);
>>>>>>        if (vcpu->arch.gs_enabled)
>>>>>> diff --git a/arch/s390/kvm/vsie.c b/arch/s390/kvm/vsie.c
>>>>>> index 9e9056cebfcf..ba83d0568bc7 100644
>>>>>> --- a/arch/s390/kvm/vsie.c
>>>>>> +++ b/arch/s390/kvm/vsie.c
>>>>>> @@ -423,6 +423,8 @@ static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
>>>>>>        break;
>>>>>>    }
>>>>>>
>>>>>> +    scb_o->cpnc = scb_s->cpnc;
>>>>>
>>>>> "This is a privileged s390x instruction that must be intercepted", how
>>>>> can the cpnc change, then, while in SIE?
>>>>>
>>>>> Apart from that LGTM.
>>>>>
>>>>
>>>> I thought shadow/unshadow was a load/store (respectively) when executing
>>>> in SIE for a level 3+ guest (where LPAR is level 1)?
>>>>
>>>> * Shadow SCB (load shadow VSIE page; originally CPNC is 0)
>>>
>>> 1. Here, you copy the cpnc from the pinned (original) SCB to the shadow SCB.
>>>
>>>>
>>>> * Execute diag318 (under SIE)
>>>
>>> 2. Here the SIE runs using the shadow SCB.
>>>
>>>>
>>>> * Unshadow SCB (store in original VSIE page; CPNC is whatever code the
>>>> guest decided to set)
>>>
>>> 3. Here you copy back the cpnc from the shadow SCB to the pinned
>>> (original) SCB.
>>>
>>>
>>> If 2. cannot modify the cpnc residing in the shadow SCB, 3. can be
>>> dropped, because the values will always match.
>>>
>>>
>>> If guest3 tries to modify the cpnc (via diag 318), we exit the SIE
>>> (intercept) in 2., return to our guest 2. guest 2 will perform the
>>> change and adapt the original SCB.
>>>
>>> (yep, it's confusing)
>>>
>>> Or did I miss anything?
>>>
>>
>> Ah, I see. So the shadowing isn't necessarily for SIE block values, but
>> for storing the register / PSW / clock states, as well as facility bits
>> for the level 3+ guests? Looking at what the
> 
> We have to forward all values the SIE has to see and copy back only what could have been changed by the SIE.
> 
>> vsie code does, that seems
>> to make sense.
>>
>> So we don't need to shadow OR unshadow the CPNC, then?
> 
> I think you have to shadow (forward the value) but not unshadow (value cannot change).
> 
> Cheers!
> 

Gotcha. Very tricky. I'll have to study on it some more. Thanks for the
info!

Take care.

>>
>> -- 
>> Regards,
>> Collin
>>
>> Stay safe and stay healthy
>>
> 


-- 
Regards,
Collin

Stay safe and stay healthy